Retrieval of live, user-broadcast video streams is an emerging problem for which the community has no established solutions. In this work, two tasks within the problem are identified, corresponding evaluation metrics proposed, and first results for challenging baselines as well as three memory-based approaches are provided.
In BMVC’16

We present Image2Emoji, a multi-modal approach for generating emoji labels from an image in a zero-shot manner.
In ACM MM’15

Technical demo for emoji search and summarization of video collections.
In ACM MM’15

Prediction of image popularity on social media using a latent ranking approach that identifies not only popular visual cues, but also unpopular visual cues.

Video event detection using bags of discriminative fragments.


Along with a fellow PhD candidate, I have organized and run the lab sessions for:

  • Visual Search Engines, 2014
  • Visual Search Engines, 2015
  • Visual Search Engines/Applied Machine Learning, 2016

These lab sessions consisted of a series of Jupyter notebooks with assignments, culminating in a class-wide machine learning competition on Kaggle. I wrote custom server software to automatically evaluate students’ code (a la Coursera), for which we received a UvA GrassRoots grant.